Reducing Hallucinations in Summarization via Reinforcement Learning with Entity Hallucination Index